Public | Automated Build

Last pushed: 5 days ago
Short Description
Short description is empty for this repo.
Full Description

ops-search - Open PHACTS search service

This is the search service for Open PHACTS, previously called IRS2.

Populates an ElasticSearch instance with JSON-LD
documents with searchable labels extracted from SPARQL queries from
a configured SPARQL service.

Exposes a Linked Data web service for searching over the indexed labels, with
content negotiation for JSON, JSON-LD, Turtle, RDF/XML etc.


License: MIT license

(c) 2014-2016 University of Manchester, UK

See LICENSE for details.

The additional Python libraries used are:


You will need Python 3 and pip, in addition to some dependencies.

In Ubuntu 14.04, this easiest achieved using:

sudo apt-get install git python3-pip libyajl2 python3-yaml python3-bottle
sudo pip3 install elasticsearch elasticsearch-dsl ijson yajl mimerender rdflib rdflib-jsonld

You will also need an ElasticSearch
installation (tested with version 1.4), with dynamic Groovy scripting

The simplest way to do this is to use the included elasticsearch Docker image:

docker run --name elasticsearch -d -p 9200:9200 openphacts/ops-search-elasticsearch

If docker ps does not show it as running then look at the logs with docker logs elasticsearch. It is possible that on Linux you may see the error max virtual memory areas vm.max_map_count [65530] is too low, increase to at least [262144]. Try setting max_map_count_value with sudo sysctl -w vm.max_map_count=262144. You can also change it in /etc/sysctl.conf but you may have to reboot.

Verify that elastic search is running at: http://localhost:9200/_search?q=alice

_Note: On OSX & Windows you may need to find the actual IP address that boot2docker is using. Try boot2docker ip and then use that when testing in a browser eg


To populate elastic search using the configuration in example.yaml, do:

python3 src/ conf/example.yaml

If you want to test the SPARQL queries without populating the elastic search indexes then add the dryrun flag and this will show you all the queries without executing them.

python src/ conf/example.yaml -d

To run the server for the API, using the same configuration, do:

python3 src/ conf/example.yaml

Updating and backing up the docker image

If the Open PHACTS Elastic Search image has been updated then you can update the image and transfer the data to the latest version.

First pull the latest version from the docker hub with docker pull openphacts/ops-search-elasticsearch.
Then stop the currently running image using docker stop elasticsearch (or whatever your image is called).
To start the latest image and to use the data from the previous version, try
docker run --name elasticsearch --volumes-from elasticsearch_old -d -p 9200:9200 openphacts/ops-search-elasticsearch where elasticsearch_old is the name for the older version (you will probably have called it something else) and elasticsearch is the docker container we want to run. You can rename docker images using docker rename elasticsearch elasticsearch_old which would rename the elasticsearch image to elasticsearch_old.

To backup the data for the elasticsearch image try docker run --rm --volumes-from elasticsearch_old -v $(pwd):/backup ubuntu tar cvf /backup/backup.tar /usr/share/elasticsearch/data. This would backup the data volume /usr/share/elasticsearch/data from the elasticsearch_old image to the local file backup.tar.


You need to create a config file similar to
example.yaml to configure the ElasticSearch data loading.

To allow external access to the API you may need to change the webservice host setting in your version of the config file to eg ''. You can also change the port where it is accessed. Note that we recommend that in production you run the API in a WSGI compatible webserver.

The list of indexes that you specify in the conf file is used to restrict searches to those branches (see below).

A description of each element of the configuration follows below:


The default configuration is:

        - host: localhost
          port: 9200

Multiple hosts can be given to address the cluster:

        - host: server1
        - host: server2
          port: 9201
        - host: server3

The address where the webservice is available can be altered:

        host: 'localhost'
        port: 8839

Additional parameters like use_ssl may be provided as supported by the ElasticSearch Python library.


A series of namespace prefixes and their URIs should be defined,
these will be used both within the generated SPARQL queries
and in the generated JSON-LD @context.

      owl: http://www.w3o.rg/2002/07/owl#

SPARQL server

The uri of a [SPARQL endpoint])( to query:

      uri: http://localhost:8890/sparql
      timeout_s: 7200 # e.g. 2 hours

TODO: Support authentication?

Common properties

The optional common_properties specifies any common properties that will
always be indexed:

      - rdfs:label

Properties MUST be given as one of:

The JSON-LD property name will be taken from the string after :, unless
that name is already used, in which case an auto-generated name is used.
You can use a property configuration to give a
better name.

Property configuration

Instead of the qname string, a property can be specified as a
nested object with the keys sparql, variable and jsonld:

      - rdfs:label
      - sparql: "dbprop:shortDescription"
        variable: "shortDesc"
        jsonld: "dc:description"

The sparql string is inserted verbatim into the query. The string
can be a qname using one of the prefixes (e.g. dc:title)
, an absolute IRI enclosed with <>
(e.g. <>), or a
SPARQL property paths
if supported by the server, e.g. foaf:knows/foaf:name.

The variable defines the SPARQL variable name that will be used in the query
(excluding the ?). This name must be unique across properties.

The jsonld defines the JSON-LD string that is used for the indexed
JSON-LD document. This MUST be given as a qname
using one of the defined prefixes.


Each key under indexes specified will (re)create the corresponding ElasticSearch index.

NOTE: Each index will be deleted before populating with new results.


Each index can contain multiple types. The below example creates the two indexes customers and staff, where the first index has two types, orgs and people, and the second only the type people.

                type: foaf:Organization
                type: foaf:Person
                type: foaf:Person

Each index-type combination can specify multiple properties for defining what to index - as detailed below:


A typical configuration for indexing multiple graphs is to have one ElasticSearch index per graph, as shown in the example above.


This specifies the name of the GRAPH to query. Note that it is always the configured sparql endpoint that is contacted, so
this URI is not retrieved.

If the graph key is missing, the default graph of the SPARQL endpoint is searched instead.


    type: foaf:Person

The rdf:type of triples to index. Typically one ElasticSearch type corresponds to one RDF type.

If type is not specified, all resources in the graph with the given properties are indexed.
In this case, all properties are required to be present.


    subclasses: owl

If the resources in the graph are only typed as subclasses of type in the graph, then specifying subclasses will modify the SPARQL query to select for subclasses instead. You can specify different mechanisms, which will generate corresponding SPARQL fragments:

subclasses: direct: will generate:

        ?uri a ?subClass .
        ?subClass rdfs:subClassOf ?type .

subclasses: owl will generate something like:

        ?uri a ?subClass .
        ?subClass a owl:Class .
        ?subClass rdfs:subClassOf+ ?type .

Note that the rdfs:subClassOf and owl:Class statements (e.g. loaded from an ontology) must be within the same graph.


The RDF properties to index:

      - dct:title
      - rdfs:description
      - skos:prefLabel
      - skos:altLabel

Properties MUST be given as one of:

Any common properties are inserted first at the top of
this list.

Within the indexed JSON documents the property will be given with the local
name of the property (e.g. title and altLabel), or with prefix_localname
if an earlier property in the property list of this index-type has a
conflicting name (e.g. dc:title dct:title would be indexed as title and

If a type was specified (match by type), then in the the generated query the
properties are individually made OPTIONAL, if not, all the specified
properties must be present in the graph (match by pattern).

API defaults

The API has a fuzziness setting of 1 to allow for simple mispellings of eg Aspirin to Asprin. The label field has also been boosted by 2 which means that it takes precedence over other fields.

To see the available indexes send a GET request to /indexes.

Request and Response

Send a GET request to /search/ with your query as the last part of the URL eg. or as the parameter query eg You can also include the branches to search for branch, the type type and the limit limit eg If you want only a list of the URIs of the hits then add an options array param which contains "uris_only". To search for multiple branches send an array listing them eg ["chebi", "chembl"]. If you are unsure of the spelling for your query you can add fuzziness by sending the option fuzzy ie options=fuzzy to the query.

The branches are:

  • chebi
  • chembl
  • uniprot
  • drugbank
  • ocrs
  • go
  • wikipathways

The types are:

  • compound
  • target
  • enzyme
  • gene

Plus all the wikipathways organism types eg human, mouse.

NOTE: depending on how your provider has chosen to index all the data you may have different branches and types. Ask them before searching.

You can also POST a query to /searchwith JSON data and the same params as for the GET query.

eg curl -H "Content-Type: application/json" -X POST -d '{"query":"Lepirudin"}' http://localhost:8839/search

Response format

Much the same as the standard Elastic Search response with the removal of the _shards element. The branch and type are added to remind you of the request params. Each hit has a highlight section which shows what triggered the response.

    "branch": "_all",
    "hits": {
        "hits": [
                "_id": "",
                "_index": "drugbank",
                "_score": 21.372051,
                "_source": {
                    "@id": "",
                    "@type": [
                    "brand_name": [
                    "label": [
                        "Lepirudin [drugbank:DB00001]"
                    "synonym": [
                        "Hirudin variant-1"
                    "title": [
                "_type": "compound",
                "highlight": {
                    "label": [
                        "<em>Lepirudin</em> [drugbank:DB00001]"
                    "title": [
        "max_score": 21.372051,
        "total": 1
    "type": "compound",
    "timed_out": false,
    "took": 22

With "uris_only" option

curl -X GET "http://localhost:8839/search?query=abc&options=uris_only&limit=10" --globoff

curl -H "Content-Type: application/json" -X POST -d '{"query":"abc", "limit": 10, "options": ["uris_only"]}' http://localhost:8839/search

    "uris": [

Here is an example searching over chebi and chembl:

curl -H "Content-Type: application/json" -X POST -d '{"query":"abc", "limit": 50, "options": ["uris_only"], "branch": ["chebi"]}' http://localhost:8839/search

and the GET version

curl -X GET "http://localhost:8839/search?query=abc&branch=chebi&branch=chembl&type=compound&limit=50&options=uris_only" --globoff

Autocomplete API

When indexing data a set of filtered labels etc using nGrams of minimum size 3 are created to allow a simple autocompleter function. Send a query param to /autocomplete. An array of possible labels is sent back.
eg for /autocomplete?query=sodium

  "value: "Sodium nitroprusside",
  "value": "Sodium carbonate"

Simple Web Application Demo

There is a web page available via the root ie http://localhost:8839 which demonstrates basic usage of the API. It provides a search box which will fetch results via GET when you type in it and a button which will POST. The ffirst 25 results are fetched and ordered with the highest hit first. Each result shows the highlight which returned the hit in bold.

Docker Pull Command
Source Repository

Comments (0)